Optimasi Model Deep Learning menggunakan Metode Gauss-Newton Terdistribusi untuk Prediksi Mutasi Sekuen Protein Spike Virus SARS-CoV-2

Jamhuri, Mohammad (2025) Optimasi Model Deep Learning menggunakan Metode Gauss-Newton Terdistribusi untuk Prediksi Mutasi Sekuen Protein Spike Virus SARS-CoV-2. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7002201001 - Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (25MB) | Request a copy

Abstract

Pandemi COVID-19 yang disebabkan oleh SARS-CoV-2 telah menjadi tantangan global, terutama dengan munculnya varian baru akibat mutasi pada protein spike. Protein ini tidak hanya berperan penting dalam infeksi virus, tetapi juga menjadi target utama dalam pengembangan vaksin. Namun, metode optimasi dalam deep learning sering kali mengalami keterbatasan dalam efisiensi komputasi dan konvergensi, terutama dalam menangani pola mutasi kompleks seperti pada varian Omicron. Oleh karena itu, penelitian ini mengembangkan metode optimasi Gauss-Newton terdistribusi untuk melatih model deep learning dalam memprediksi mutasi protein spike SARS-CoV-2 secara lebih akurat dan efisien. Hasil penelitian menunjukkan bahwa metode Gauss-Newton terdistribusi mampu mengurangi waktu komputasi hingga 30% dibandingkan metode optimasi seperti SGD, AdaGrad, SGDM, dan Adam. Model deep learning yang digunakan, yaitu MLP, CNN, dan LSTM, mencapai akurasi prediksi mutasi hingga 99% pada varian Alpha, Beta, Delta, dan Gamma. Namun, pada varian Omicron, yang memiliki kompleksitas mutasi lebih tinggi dan ketersediaan data lebih terbatas, terjadi sedikit penurunan akurasi sekitar 1–2%, yang menunjukkan tantangan dalam menangani evolusi virus yang lebih dinamis. Kontribusi utama penelitian ini meliputi penerapan Gauss-Newton terdistribusi untuk meningkatkan efisiensi optimasi deep learning dalam klasifikasi dan prediksi mutasi, transformasi multivariable time series forecasting menjadi klasifikasi multilabel, serta pengembangan pendekatan diagonal untuk efisiensi komputasi skala besar. Secara praktis, metode ini dapat mendukung sistem pemantauan evolusi SARS-CoV-2 secara real-time, meningkatkan efektivitas pengawasan varian virus, serta mempercepat adaptasi strategi pengembangan vaksin dan mitigasi pandemi. Selain itu, pendekatan ini dapat diperluas untuk penelitian virus RNA lain, sehingga berkontribusi dalam pengawasan penyakit infeksi secara global.
================================================================================================================================
The COVID-19 pandemic, caused by SARS-CoV-2, has posed a global challenge, particularly with the emergence of new variants due to mutations in the spike protein. This protein plays a crucial role in viral infection and serves as the primary target for vaccine development. However, optimization methods in deep learning often face computational inefficiencies and convergence issues, especially when handling complex mutation patterns, such as those in the Omicron variant. To address this challenge, this study develops a distributed Gauss-Newton optimization method to train deep learning models for predicting spike protein mutations in SARS-CoV-2 more accurately and efficiently. The results demonstrate that the distributed Gauss-Newton method reduces computational time by up to 30% compared to optimization methods such as SGD, AdaGrad, SGDM, and Adam. The deep learning models used, including MLP, CNN, and LSTM, achieved mutation prediction accuracy of up to 99% for Alpha, Beta, Delta, and Gamma variants. However, for the Omicron variant, which exhibits higher mutation complexity and limited data availability, a slight accuracy drop of around 1–2% was observed, highlighting the challenges of handling more dynamic viral evolution. The key contributions of this study include the application of distributed Gauss- Newton to improve deep learning optimization efficiency in classification and mutation prediction, the transformation of multivariable time series forecasting into a multilabel classification problem, and the development of a diagonal approach for large-scale computational efficiency. In practical terms, this method can support real-time monitoring of SARS-CoV-2 evolution, enhance variant surveillance effec- tiveness, and accelerate the adaptation of vaccine development and pandemic mitigation strategies. Furthermore, this approach can be extended to research on other RNA viruses, contributing to global infectious disease surveillance

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Gauss-Newton, Deep Learning, Spike Protein, Mutation Prediction, SARS-CoV-2, Gauss-Newton, Deep Learning, Protein Spike, Prediksi Mutasi, SARS-CoV-2
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44002-(S3) PhD Thesis
Depositing User:	Mohammad Jamhuri
Date Deposited:	02 Feb 2025 14:04
Last Modified:	02 Feb 2025 14:04
URI:	http://repository.its.ac.id/id/eprint/117554

Actions (login required)

View Item